Big Tech firms are building their own AI chips to reduce dependence on Nvidia

Nvidia is confronting its most serious challenge yet, as its biggest clients start cutting into the same market that built its trillion‑dollar empire.

According to Yahoo Finance, tech giants including Amazon, Google, Meta, Microsoft, and OpenAI are developing their own AI chips to reduce dependence on Nvidia’s hardware.

The move could slash Nvidia’s margins, as these companies prepare to rely less on its expensive GPUs for training large AI models and running massive data centers.

OpenAI, which rents Nvidia chips through Microsoft and CoreWeave, said it has started designing its own custom chips in partnership with Broadcom. Meta announced in late September that it will acquire Rivos, a chip startup that fits into its broader plan to expand internal chipmaking.

Over at Amazon, its huge data center effort called Project Rainier is already “well underway,” with hundreds of thousands of Trainium2 chips powering AI workloads from Anthropic. Analysts said demand for Amazon’s chips deployed in those centers has risen fast, hinting at how much less reliant these companies may soon become on Nvidia.

Big Tech turns inward with custom chip designs

While Nvidia’s GPUs still dominate the AI market, cloud providers are now designing their own chips with Broadcom and Marvell Technology.

These custom processors are cheaper, tuned for their software, and make it easier to control performance costs. The chips aren’t sold to outsiders like Nvidia’s GPUs are, they’re used internally to run AI systems, and offered to cloud clients at lower prices.

In a June research note, JPMorgan projected that chips from Google, Amazon, Meta, and OpenAI will make up 45% of the AI‑chip market by 2028, compared with 37% in 2024 and 40% in 2025. The rest of the market will remain with GPU producers such as Nvidia and AMD.

Jay Goldberg, an analyst at Seaport Research, said the hyperscalers are building custom silicon because “they don’t want to be stuck behind an Nvidia monopoly.” He added that Nvidia now has to “compete with its customers.”

That’s already happening. Google reportedly began selling its TPUs, or tensor processing units, to a cloud provider in September, a decision that puts it in direct competition with Nvidia.

Gil Luria, an analyst at DA Davidson, estimated Google’s TPU and DeepMind units could be worth $900 billion, calling them “arguably one of Alphabet’s most valuable businesses.” He wrote that Google’s chips “remain the best alternative to Nvidia, with the gap between the two closing significantly over the past nine to twelve months.”

Goldberg predicted “a lot of activity around custom silicon” by 2026, reflecting conversations throughout the AI chip supply chain. Analysts said Big Tech companies are progressing at different speeds.

Google began developing TPUs more than a decade ago and remains the leader. Amazon entered the space a year after Google launched its first TPU, buying Annapurna Labs in 2015 and releasing Trainium in 2020. Microsoft, which only launched its Maia AI chip in 2023, still lags its peers.

Analysts warn of slower growth for Nvidia

Developers often prefer Nvidia’s GPUs because of the software stack that comes with them, but analysts said the competition will still eat into profits. David Nicholson, from Futurum Group, said the margin risk is real.

“Over time, the margins that Nvidia can command right now get degraded,” said David. “It will be sort of death by a thousand cuts because you have all of these different custom silicon accelerators that exist because there’s such an opportunity.”

When asked about this in a September podcast, Jensen Huang downplayed the threat. The Nvidia CEO said, “We’re the only company in the world today that builds all of the chips inside an AI infrastructure.”

He argued that while rivals are building single chips, Nvidia produces complete systems combining Blackwell GPUs, Arm‑based CPUs, and networking units that work together across entire racks.

Some on Wall Street don’t share the same fear. Vivek Arya of Bank of America and Luria both said the rise of custom chips “doesn’t matter.” Arya said Nvidia keeps expanding the total market, noting that the company invested $47 billion into AI and “neocloud” startups from 2020 through September 2025, according to PitchBook data.

Luria added, “The growth and demand is so substantial. We’re going to need a lot more compute, and the AI models are getting more useful, which means the pie is gonna get a lot bigger.”

Luria said Nvidia may not grow as fast as the market itself, but will still expand as overall demand rises. Still, Goldberg warned that designing chips isn’t easy. “The drawback of doing your own silicon, though, is that it’s hard,” he said. “I think ultimately what will happen is not all of them will succeed.”

Want your project in front of crypto’s top minds? Feature it in our next industry report, where data meets impact.